Medical informed machine learning: A scoping review and future research directions

Area of Science:

• Medical Artificial Intelligence
• Machine Learning
• Domain Knowledge Integration

Background:

• Machine learning (ML) in medicine faces challenges like limited explainability, data scarcity, and robustness issues.
• Current informed machine learning (IML) approaches are often problem-specific, limiting broader application.
• Integrating domain knowledge (DK) with ML offers a promising avenue to address these limitations.

Purpose of the Study:

• To conduct a scoping literature review of informed machine learning (IML) in medicine.
• To analyze the current status of IML in the medical field.
• To identify trends, challenges, and future research directions in medical IML.

Main Methods:

• A scoping literature review was performed using an existing taxonomy.
• 177 relevant papers were identified and analyzed.
• Analysis focused on the type of domain knowledge (DK) used, the machine learning (ML) models implemented, and the motivations for employing IML.

Main Results:

• Expert knowledge and medical image data are extensively utilized in medical IML.
• A significant number of IML approaches are tailored to specific medical problems.
• The review identified key trends in DK integration and ML model selection within medical IML.

Conclusions:

• This review provides a comprehensive overview of the current landscape of IML in medicine.
• It highlights the critical role of expert knowledge and image data in advancing medical AI.
• The study offers valuable insights and five future research directions to guide the development of more robust and explainable medical IML systems.